Your Car Knows You're Getting Sick

In 2023, I joined Fisker before they even shipped a single car. I'd been building software for years, and I wanted to work on something physical — something that actually interacts with the world. But the real reason I joined wasn't electric vehicles.

Luke Miller, a friend of mine at LSU had published research showing that vehicle signal data — steering patterns, braking behavior, acceleration habits — could be used to detect cognitive decline in the driver. Not with special medical equipment. Not with a clinical visit. Just from how you drive. I read that, and wanted to build a real-time solution.

Your car could flag early signs of neurological deterioration before your doctor suspects anything. The sensors already exist. The data is already being generated. It just wasn't being made available.

Fisker was building on Magna's brand new EV platform and already wanted to pull signals off the car for machine learning — performance tuning, predictive maintenance, and driver assist. But they were small enough that the ML direction wasn't completely set yet.

Autonomous driving from Waymo, Applied Intuition, and Zoox understandably dominates the conversation around ML and vehicles. But there's also room for a novel application of the same data: treating a car as a health product.

We built the data pipeline — processing 2 million messages a second off the cars, preparing signal data for training, storing 8TB of signals in parquet files a day. The most notable win was the model to gain the last 10 miles of range on the Ocean, surpassing the Tesla Model Y. The infrastructure for a health application was there and we were accumulating data fast.

Then Fisker imploded — somewhat spectacularly. Once the factory line shut down, a health application built on vehicle signal data wasn't going to happen there. But I'm really happy I got a shot at building something unique.